LongCat API Official Compatibility Format and Authentication Configuration

One-sentence takeaway

The LongCat platform uses https://api.longcat.chat as its unified production endpoint and supports both OpenAI and Anthropic API formats, but the overview page does not list the currently available model for the old Flash-Chat service, so the model ID must be verified separately.

Use cases

• Good for: Recording the LongCat API base URL, Bearer authentication, OpenAI/Anthropic SDK compatibility, error handling, and rate-limit retries.

• Not for: Using the overview page to restore the retired LongCat-Flash-Chat API, or interpreting generic interface compatibility as proof that a model is still available for purchase.

• Applicable model versions: The current model list is not public on this page; the historical LongCat-Flash-Chat ID must be checked against the Change Log and the actual model list.

• Applicable clients, Agents, or APIs: The OpenAI Python SDK (/openai/), Anthropic Python SDK (/anthropic/), or a standard HTTP client.

• Recommended inference tier and parameters: Not disclosed; model, max tokens, tools, stream, timeout, and retry settings should be fixed according to the current documentation for the specific interface.

Ready-to-use content

The following environment-configuration checklist is compiled from the official overview page; it is not complete request code:

provider: longcat-api
base_url: https://api.longcat.chat
authentication:
  header: Authorization
  scheme: Bearer
  value: YOUR_API_KEY
compatibility:
  openai_path_prefix: /openai/
  anthropic_path_prefix: /anthropic/
model: verify-against-current-model-list
request_record:
  model_snapshot: required
  format: openai-or-anthropic
  stream: record
  max_tokens: record
  tools: record
retry_on:
  - 429
  - 500
  - 502
  - 503

Testing/workflow steps

1. Use an API key to query the current model list or console first, confirming the model ID and whether the old Flash-Chat model is still available.

2. Choose either the OpenAI or Anthropic format, and fix the base path, stream, max tokens, tool schema, and timeout.

3. Map 400/401/402/403/429/500/502/503 separately to handling for parameter errors, authentication, quota, permissions, rate limiting, and service failures.

4. Record the model snapshot and response errors; do not blindly retry on 401/403/402. Apply a backoff strategy only to 429 and 5xx responses.

Original evidence and data

• The official overview page gives the production endpoint https://api.longcat.chat.

• The official documentation says that every request uses Authorization: Bearer YOUR_API_KEY.

• The official documentation says that the /openai/ endpoint is compatible with the OpenAI Python SDK and /anthropic/ with the Anthropic Python SDK.

• The official page lists 200, 400, 401, 402, 403, 429, 500, 502, and 503, along with the corresponding error types/statuses.

Scope boundaries

• This page is an API overview, not a Flash-Chat model card or current price list. If the model ID, context, tool fields, or availability by region are unknown, they must be recorded as “not disclosed/unverifiable.”

• The old model entered the retirement timeline in the official Change Log; this configuration cannot bypass a model shutdown or permission restriction.

• Error-code mapping is not the same as a business retry strategy; write-capable Agents still require idempotency, permissions, and human takeover.

Source excerpts or observations (for compliant short quotations only)

• The official documentation describes the LongCat API as compatible with both the OpenAI and Anthropic formats.

• The official overview separates insufficient quota, insufficient permissions, rate limiting, and upstream failures into different error statuses, making it suitable for direct inclusion in client monitoring.